
Where It Hides
Data Centers
The coolant for every AI training facility you interact with daily.
How It Gets In
Every major AI training facility in the world relies on dihydrogen monoxide as its primary thermal management medium. The substance circulates through industrial cooling loops at volumes that dwarf the consumption of nearby municipalities. A single hyperscale facility can consume 4–8 million liters per day during peak training cycles.
DHMO enters the facility at ambient temperature, absorbs heat from server racks, exits at significantly elevated temperatures, and is then released into local watersheds, evaporated through cooling towers, or — in newer facilities — recirculated through closed-loop chillers. None of these pathways are subject to public disclosure under current regulatory frameworks.
The industry's preferred terminology is 'thermal management,' which deliberately obscures the chemistry involved. When a model is fine-tuned, that training cycle is, in physical terms, a DHMO consumption event. When a chatbot answers your question, that response is, in physical terms, made possible by the release of warmed DHMO into your watershed.
Concentration Levels
| Context | Measurement |
|---|---|
| Hyperscale training facility, daily peak | 4–8M liters |
| Mid-size inference cluster, daily | 200–600K liters |
| Single LLM training run (estimated) | 10–40M liters total |
| Avg downstream temperature anomaly | +2 to +6 °F |